Back to AIBriefs
AnalysisPolicyAI Models

BiasGRPO stabilizes LLM bias mitigation with group-relative optimization

Arxiv CS.AI (top papers)

The paper introduces BiasGRPO, which uses group-relative policy optimization to stabilize bias mitigation in LLMs under high-variance reward conditions. Unlike verifiable tasks, bias mitigation lacks a single ground truth, making alignment challenging.

·
7 days ago
BiasGRPO stabilizes LLM bias mitigation with group-relative optimization — AIBriefs